Automatic Contrast Enhancement using Ensemble Empirical Mode Decomposition
Date Issued
2011
Date
2011
Author(s)
Lin, Shang-Ching
Abstract
Ultrasound nonlinear contrast imaging using microbubble-based contrast agents has been widely investigated. However, the degree of contrast enhancement is often limited by overlap between the spectra of the tissue and microbubble nonlinear responses, which makes it difficult to separate them. The use of ensemble empirical mode decomposition (EEMD) in the Hilbert-Huang transform (HHT) was previously explored with the aim of alleviating this problem. The HHT is designed for analyzing nonlinear and nonstationary data, whereas EEMD is a method associated with the HHT that allows decomposition of data into a finite number of intrinsic mode functions (IMFs). It was found that the contrast can be effectively improved in certain IMFs, but manual selection of appropriate IMFs is still required. This prompted the present study to test the hypothesis that the contrast can be enhanced without requiring manual selection by summing appropriately weighted IMFs and demodulating the signal at appropriate frequencies. That is, a data-driven mechanism for automatically determining weights and demodulation frequencies was derived and tested. Users only have to specify the microbubble distribution in the training data set, and the contrasts in testing data sets can be improved. Phantom results show that an overall contrast enhancement of up to 12.5 dB can be achieved. A fused-image representation that simultaneously displays the conventional B-mode image and the new contrast mode image is also presented. The proposed method outperforms second-harmonic imaging significantly, but is only slightly better than subharmonic imaging on experimental data. However, there is a limitation that the imaging setups should be identical for obtaining training and testing data. Though there are other means to determine the weights, as long as they are determined through a training process, the contrast improvement and the reliability of the results will mainly depend on the size of the training data set. Finally, in general the proposed method demands more computations than conventional methods. Hence, future studies will not only tempt to apply the method to other imaging configurations and clinical data, but also seek for a set of computational parameters or utilize other algorithms derived from ensemble empirical decomposition (EMD) to better balance computational complexity and contrast improvement.
Subjects
ultrasound contrast imaging
microbubble
nonlinear imaging
empirical mode decomposition (EMD)
ensemble empirical mode decomposition (EEMD)
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-100-R98945011-1.pdf
Size
23.32 KB
Format
Adobe PDF
Checksum
(MD5):ee67d303a8f6d71ab161d3f307f57540